Functional Network Connectivity Dynamics in the Human Fetal Brain

1824 

Abstract Submission 

Tanya Bhatia¹, Lanxin Ji², Ellyn Kennelly³, Amyn Majbri¹, Mark Duffy², Iris Menu², Moriah Thomason⁴

¹New York University Medical Center, New York, NY, ²NYU Langone Health, New York, NY, ³Wayne State University, Detroit, MI, ⁴NYU Langone Medical Center, New York, NY

Tanya Bhatia  
New York University Medical Center
New York, NY

Lanxin Ji  
NYU Langone Health
New York, NY

Ellyn Kennelly  
Wayne State University
Detroit, MI

Amyn Majbri  
New York University Medical Center
New York, NY

Mark Duffy, MS  
NYU Langone Health
New York, NY

Iris Menu  
NYU Langone Health
New York, NY

Moriah Thomason  
NYU Langone Medical Center
New York, NY

The human brain undergoes dramatic transformations during the late fetal stage, with the emergence and organization of brain networks. Fetal MRI has made it possible to map the functional connectivity of the brain during this critical developmental period [9]. Recent literature suggests that dynamic alterations in human brain functional connectivity during an MRI scan may be indicative of large-scale functional capacity and risk for abnormalities and disease [2,4]. However, no studies to date have explored temporally-driven connectivity dynamics in the fetal brain.

Imaging data were acquired from 101 fetuses (39 females), gestational age 24.00-37.86 weeks (31.19 ± 3.71), who were enrolled in the Perinatal Imaging of Neural Connectivity study. Functional MRI were acquired using a 3T Siemens Verio 70 cm open-bore system with an abdominal 4-channel Siemens Flex coil, with the following gradient multi-echo planar imaging sequence: TR = 2000 ms; TEs = 18, 34.06, 50.12ms; flip-angle: 80-degree, slice-gap: none; voxel-size: 3.4 x 3.4 x 4 mm3. Preprocessing included brain segmentation with deep learning, motion estimation and censoring with FSL FLIRT. Participants with fewer than 105 low-motion frames were excluded. Optimal combination across echoes, normalization to standard space (GA = 32 weeks), smoothing, and ICA and CompCor denoising were subsequently completed.

Spatial group independent component analysis (ICA) was implemented in the Group ICA of functional MRI Toolbox [6] to extract functional brain networks. Each fetus's time course profile was based on the Infomax [1] algorithm. The number of ICs was chosen to be 40 based on MDL criterion [3].

34 of the 40 components were identified as non-noise and grouped into 9 functional networks: cerebellum, temporoparietal, frontoinsular, frontal pole, DMN, temporal regions, motor, subgenual, and visual (Fig 1). The time courses of 34 ICs were then detrended, despiked, and filtered with a high frequency cut-off of 0.15 Hz [4]. Pair-wise Pearson's correlations between the ICs were then calculated and z-transformed to form the static functional connectivity matrix.

Dynamic functional connectivity analysis was then performed using the dynamic functional network connectivity (DFNC) tool in GIFT. A sliding-window approach was first used to investigate time-varying changes in functional connectivity within the 34 IC networks. Time courses were segmented into a 30-repetition time (TR) window, with a step-wise TR of 2 seconds. K-means clustering (k=3) was then conducted to explore recurring states (i.e. functional connectivity patterns). The correlation of the window count of each state was then correlated with fetal sex using point-biserial correlation, and with gestational age at the time of scan using Pearson's correlation.

Of the 3 FC states, 2 of them were dominant – 42% in State 1 and 54% in State 2. State 3 was disregarded due to its low frequency of subjects. State 1 was slightly more connected than State 2, both within- and between-networks (Fig 2A). For example, the within-network connectivity of motor, visual, default mode network (DMN), and cerebellar regions is higher in State 1 than State 2. The connectivity between the DMN and visual regions is also higher in State 1. Subjects tended to dwell in State 2 the longest, followed by State 1 (Fig 2B). Subjects tended to dwell in a given state rather than transitioning states, with just 0.4 +/- 0.7 transitions on average (Fig 2C). No significant correlations were found between the count of each state with fetal sex or gestational age at scan.

This exploration of fetal functional connectivity (FC) dynamics revealed two main states with similar FC profiles, and there was minimal state transition, implying low FC dynamics in the fetus. Further work should investigate if these late fetal states persist in varying samples and predictive factors of state characteristics.

Normal Brain Development: Fetus to Adolescence ²

fMRI Connectivity and Network Modeling ¹

Task-Independent and Resting-State Analysis

Normal Development

Computational Neuroscience

Data analysis

Development

FUNCTIONAL MRI

Machine Learning